AI-Driven Analysis of Users' reactions to deep-fake imagery in Archaeological Misinformation

Fake news and deep-fakes are introducing information disorder even in the digital humanities, where misleading narratives and manipulated images can distort the collective perception in the context of history and art. This study proposes a depth analysis of the phenomenon analyzing the users' reactions to elements of misinformation related to cultural field. Using an approach based on Deep Learning and Large Language Models (LLMs), we classified and validated comments collected from social media to estimate the level of acceptance or skepticism concerning different fake news and deep-fakes. In addition classification techniques were tested, comparing methodologies based on deterministic assignment and strategies with uncertainty thresholds to improve the accuracy of the analysis. The results, presented and discussed in the paper, confirm the widespread misinformation in digital platforms. The performance of the developed model shows good classification ability, however, open challenges remain, including the difficulty in recognizing sarcasm and linguistic ambiguities, aspects that may still affect the accuracy of classification.